maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

Implement async_work_group_strided_copy builtin v2 

This is a simple implementation which just copies data synchronously.

v2:
  - Use size_t.

llvm-svn: 219007
This commit is contained in:
Tom Stellard 2014-10-03 19:49:37 +00:00
parent b5064f79ef
commit ed5bbfdb1b
6 changed files with 66 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
libclc/generic/include/clc/async/async_work_group_strided_copy.h Normal file
View File

@ -0,0 +1,15 @@
#define __CLC_DST_ADDR_SPACE local
#define __CLC_SRC_ADDR_SPACE global
#define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
#include <clc/async/gentype.inc>
#undef __CLC_DST_ADDR_SPACE
#undef __CLC_SRC_ADDR_SPACE
#undef __CLC_BODY
#define __CLC_DST_ADDR_SPACE global
#define __CLC_SRC_ADDR_SPACE local
#define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
#include <clc/async/gentype.inc>
#undef __CLC_DST_ADDR_SPACE
#undef __CLC_SRC_ADDR_SPACE
#undef __CLC_BODY

6
libclc/generic/include/clc/async/async_work_group_strided_copy.inc Normal file
View File

@ -0,0 +1,6 @@
_CLC_OVERLOAD _CLC_DECL event_t async_work_group_strided_copy(
__CLC_DST_ADDR_SPACE __CLC_GENTYPE *dst,
const __CLC_SRC_ADDR_SPACE __CLC_GENTYPE *src,
size_t num_gentypes,
size_t stride,
event_t event);

1
libclc/generic/include/clc/clc.h
View File

@ -137,6 +137,7 @@
#include <clc/synchronization/barrier.h>
/* 6.11.10 Async Copy and Prefetch Functions */
#include <clc/async/async_work_group_strided_copy.h>
#include <clc/async/prefetch.h>
#include <clc/async/wait_group_events.h>

1
libclc/generic/lib/SOURCES
View File

@ -1,3 +1,4 @@
async/async_work_group_strided_copy.cl
async/prefetch.cl
async/wait_group_events.cl
atomic/atomic_xchg.cl

9
libclc/generic/lib/async/async_work_group_strided_copy.cl Normal file
View File

@ -0,0 +1,9 @@
#include <clc/clc.h>
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
#endif
#define __CLC_BODY <async_work_group_strided_copy.inc>
#include <clc/async/gentype.inc>
#undef __CLC_BODY

34
libclc/generic/lib/async/async_work_group_strided_copy.inc Normal file
View File

@ -0,0 +1,34 @@
#define STRIDED_COPY(dst, src, num_gentypes, dst_stride, src_stride) \
size_t size = get_local_size(0) * get_local_size(1) * get_local_size(2); \
size_t id = (get_local_size(1) * get_local_size(2) * get_local_id(0)) + \
(get_local_size(2) * get_local_id(1)) + \
get_local_id(2); \
size_t i; \
\
for (i = id; i < num_gentypes; i += size) { \
dst[i * dst_stride] = src[i * src_stride]; \
}
_CLC_OVERLOAD _CLC_DEF event_t async_work_group_strided_copy(
local __CLC_GENTYPE *dst,
const global __CLC_GENTYPE *src,
size_t num_gentypes,
size_t src_stride,
event_t event) {
STRIDED_COPY(dst, src, num_gentypes, 1, src_stride);
return event;
}
_CLC_OVERLOAD _CLC_DEF event_t async_work_group_strided_copy(
global __CLC_GENTYPE *dst,
const local __CLC_GENTYPE *src,
size_t num_gentypes,
size_t dst_stride,
event_t event) {
STRIDED_COPY(dst, src, num_gentypes, dst_stride, 1);
return event;
}